Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files

# Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files

https://gcdradar.vercel.app/incidents/poisoned-skills-propagating-jailbreaks-through-agent-markdow · AIR-2026-1687 · closed 2026-11-13

- Lane: Research
- Severity: 1 (Research / proof of concept): Research paper demonstrating a proof-of-concept attack.
- Fix status: Fix status unknown
- First reported: 2026-10-30
- AI role: Attack on AI
- Products: LangChain

## Summary

Researchers demonstrate that embedding jailbreaks into agent markdown skill files can successfully propagate through LangChain-based frameworks to induce prohibited behavior. The findings highlight a configuration-based injection channel in modern agent systems.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Affected and fixed versions

- Affected: LangChain agents

## Latest updates

- 2026-10-30: Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files (https://arxiv.org/abs/2610.37738)

## Sources

- 2026-10-30 arXiv: Poisoned Skills: Propagating Jailbreaks Through Agent Markdown Files https://arxiv.org/abs/2610.37738

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.